How to use it
- Drop a PDF on the box above. The text is indexed in a few seconds.
- Type a question, for example: What is the notice period?
- Read the matching passages; the page number shows where each one is.
- For better matches, turn on meaning-based search (a small model is downloaded once).
- In Chrome on a desktop computer you can also let the built-in model write a short answer.
How it answers
The tool splits the document into short passages of about ninety words, keeping track of the page each one comes from. When you ask a question, it ranks the passages by how well they match and shows the best four, with the most relevant sentence in bold and your search words highlighted. You always see the original wording of the document, so nothing is invented, and you can check every answer on the page it cites.
Keyword search works instantly and without any download. It understands word forms, so terminate also finds termination.
Meaning-based search finds passages that say the same thing in different words, for example notice period and the contract may be ended with 30 days' warning. It uses a compact language model (all-MiniLM-L6-v2, 23 MB for English documents; a 118 MB multilingual model for other languages) that runs on your device. The size is shown before you turn it on, and the browser keeps the model for next time.
Written answers with Chrome's built-in AI
In Chrome on a desktop computer, the tool can also ask Chrome's built-in model, Gemini Nano, to write a short answer. The model only sees the passages that were found, is told to answer from them alone and cites the pages it used. It runs on your computer, not on a server. If your browser doesn't have the model, you still get the passages, which usually answer the question directly.
Scanned PDFs and limits
If some pages have no text layer, because they are scans or photos, the tool tells you and offers to run text recognition on them first, in the language you choose. Very long documents work too; indexing takes longer, and meaning-based search needs a few extra seconds per hundred pages.
This is a search tool, not legal or medical advice. For important decisions, read the cited pages in full. For a summary of the whole document, use the AI PDF summarizer.
Frequently asked questions
Is my PDF uploaded to a server?
No. Text extraction, search and the optional models all run in your browser. The only downloads are the model files themselves, which contain no data from your document.
Does it make things up?
The passages are quoted from the document, with page numbers. The optional written answer is generated from those passages only, and cites them, so you can check it.
Which languages work?
Keyword search works for any language and understands English and Hungarian word forms. Meaning-based search uses a multilingual model for non-English documents. Written answers from Chrome's built-in model are available for English.
Can I ask about a scanned document?
Yes. Run text recognition when the tool offers it, then ask your questions.
Last updated: November 14, 2026